Improved PSO based K-Means Clustering Applied to Fault Signals Diagnosis

In this paper, we attempt to detect and diagnose anomalous from fault signals in industrial processes using K-means clustering technique with Particle Swarm Optimization (PSO)algorithm. In addition to healthy (normal)engines, we consider three types of engine faults: joints problem, faulty bearings and mechanical loosening. Data are characterized by six discriminant variables and the aim is to classify them into four classes. To carry out this analysis, three algorithms are used. First, conventional K-means is applied to the considered data. In a second approach, K-means is combined with the PSO technique. The improvement proposed in the third algorithm concerns the use of PSO based K-means with Global K-means initialization. The performances of our approach are improved and the results compare favorably with the use of the traditional K-means algorithm.